Nomic Embed v2
Free.ai (self-hosted)
·
embeddings
·
~100 tokens per call
Nomic Embed v2 ke an embedding model e hahiloeng ke Nomic AI. E ka ba e le Retrieval augmented generation with flexible vector sizes.. E na le li-Free.ai GPUs tse nang le li-server tse 100 tse nang le li-server tse 100 tse nang le li-server tse 100. E lokollotsoe ka tlas'a Apache 2.0 - ho sebelisoa ka khoebo ho lumelloa ka Free.ai.
Lipotso tse botsoang khafetsa
Nomic Embed v2 e fetolela mongolo ho ea ho vector e thata (lista ea li-floats) e leng e etsang hore ho be le tlhaloso. E sebelise ho batla ka semantic, ho kopanya, ho khothaletsa, ho fumana ho eketseha ho theha (RAG), le mosebetsi ofe kapa ofe moo "na mongolo ona o tšoana le mongolo oo" o leng bohlokoa.
Li-dimensions tse tloaelehileng ke 384, 768, 1024, kapa 1536 ho latela mofuta. BGE-M3 e fana ka 1024-dim; OpenAI Ada e fana ka 1536. Likarabo tsa API li kenyelletsa boholo ka hona vector DB ea hau e khetha index e nepahetseng.
Litšoantšo tse ncha tsa ho kenya (ho akarelletsa le likhetho tse ngata ho Free.ai) li koetliselitsoe ka lipuo tse fetang 100. Cross-language retrieval works - search in English, match documents in Spanish.
512 ho 8,192 tokens ho latela mofuta. Li-input tse telele li tlosoa - chunk litokomane tse telele ho liparagrape pele ho embed.
Nomic Embed v2 e sebetsa ka GPUs ea rona le ke e 'ngoe ea lisebelisoa tse theko e tlaase - ka li-token tse ka bang 100 ka ho ngola tse tsoang ho pool ea hau ea mahala ka letsatsi. $ 5 = 200K tokens.
Ea - POST lethathamo la lihlooho ho /v1/embeddings/ le Nomic Embed v2 e khutlisa lethathamo la li-vectors ka tsela e ts'oanang. Boholo ba batch ho fihlela ho 2,048 ka kopo.
L2-normalized ka ho tloaeleha - cosine similarity = dot product. Pass `normalize=false` if you want raw vectors for a different distance metric.
Any - Pinecone, Weaviate, Qdrant, Chroma, pgvector, FAISS, LanceDB. Nomic Embed v2 e khutlisa li-floats tsa JSON tse tloaelehileng; DB ha e bone model.
Ea - POST ho /v1/embeddings/ le model="Nomic Embed v2". OpenAI-compatible response shape, so existing client libraries work unchanged. /api/ has the full reference.
Li-model tse sebete li boloka mongolo oa hau ho GPUs tsa rona'me li o tlohela ka mor'a hore ho khutle ho ngola. Premium e tsamaea ka DPA. Re sa koetlise ka ho kenya ha hao.
Sub-100ms bakeng sa mongolo o khuts'oane ka ho arolelana, 100-500ms ka premium. Li-call tsa batch li phahame ka ho lekana - 1,000 chunks li phetheloa ka metsotsoana ea 2-10.
Yeah - Free.ai e fana ka ts'ebeliso ea khoebo ea ho embeddings. E theha ho batla tlhahiso, RAG pipelines, li-systems tsa ho khothaletsoa ntle le royalty ea per-vector.